Thank for your advice, Serge! My code would indeed generate a lot of
denormals.
Unfortunately, in my case turning JavaMode off does not seem to
correlate with the resulting speed. During testing, I had all for
combinations of (fast / slow) execution and (turning JavaMode off / not
tinkering with JavaMode).
So I'm still looking for a solution...
On 29 Nov 2007, at 10:49, Serge Cohen wrote:
I had the same problem going from 10.3 to 10.4 ...
In this case the problem was due to Accelerate not using "flush
denormals" anymore, but providing an IEEE compatible treatment of
denormals (which basically means that computation is made by the
scalar fpu). In other words, Accelerate would always turn the VE's
java Mode to ON, and it would NOT turn the mode back to the original
one.
In my case, I could get back to close the previous performance by just
explicitly turning java mode off after each call to accelerate (as the
rest of my code was hitting denormals more often than the BLAS calls).
The difference between your case and mine might be that in my case, the
BLAS call itself is slow, while performance after the call seems
unaffected.
In this latter one, the text mentions that by default G5 (I presume
also G4 ?) have the default of handling denormals to OFF. Indeed this
is a bit misleading in the sense that this default will be changed as
soon as you call BLAS/LAPACK from Accelerate (as much as I
experienced).
Indeed I would read the Apple documentation as JavaMode always being
off on PPC and on on Intel.
Thanks,
Tim
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Scitech mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
http://lists.apple.com/mailman/options/scitech/email@hidden
